To avoid capturing donor-specific changes, we applied correlation and differential expression analyses for each of the six brain donors separately, and effect sizes were then combined by meta-analysis (metafor R-package 2.0). A random effects model was applied which assumes that each brain is considered to be from a larger population of brains and therefore takes the within-brain and between-brain variance into account. The between-brain variance (tau2) was estimated with the Dersimonian-Delaird model. Variances and confidence intervals were obtained using the escalc-function. Correlations were Fisher-transformed (z = arctanh(r)) to obtain summary estimates, which were then back-transformed to correlation values ranging between −1 and +1. The significance of summary effect sizes (correlations and FCs) was assessed through a two-sided t-test with 5 degrees of freedom (H0:FC = 0). P-values were BH-corrected for all 20,017 genes. The weights used in the meta-analysis are based on the non-pooled expression variance in R1–R6.
